freddyc

I've been trying to get background subtraction using jit.openni and Kinect but I haven't managed to achieve it so far. I can do it with jit.freenect.grab but for some reason, I can't achieve it with jit.openni. I need to use jit.openni because of the skeleton tracking. Any ideas ?

background-subtraction-with-jit-openni-kinect

There are a couple of posts on the forum about this. Basically what you need to do is take the usermap output and evaluate wether its values are greater than zero, and then multiply the result by the depthmap pixels. The usermap contains a matrix of all zeros except for where it detects human shapes. Then wherever a the first detected user is will be all 1's. A second detected user would have all 2's and so on up to six users (i think).

The masking could be done with a jit.expr with a expression such as: in[1]

This approach works much better than background subtraction, but it only works with human shapes, with a mostly frontal view. If you want to mask based on distance or you are mounting the kinect above looking down, you will likely have to take a different approach.

There are a couple of posts on the forum about this.  Basically what you need to do is take the usermap output and evaluate wether its values are greater than zero, and then multiply the result by the depthmap pixels. The usermap contains a matrix of all zeros except for where it detects human shapes.  Then wherever a the first detected user is will be all 1's.  A second detected user would have all 2's and so on up to six users (i think).  

This approach works much better than background subtraction, but it only works with human shapes, with a mostly frontal view.  If you want to mask based on distance or you are mounting the kinect above looking down, you will likely have to take a different approach.


This approach works much better than background subtraction, but it only works with human shapes with a mostly frontal view. If you want to mask based on distance or you are mounting the kinect above looking down you will likely have to take a different approach.

This approach works much better than background subtraction, but it only works with human shapes with a mostly frontal view.  If you want to mask based on distance or you are mounting the kinect above looking down you will likely have to take a different approach.


I'm intending place the kinect above looking down so i'm not sure if this will be the right solution for me. I will try it anyway.

I tried a method that seems to be working fine with jit.freenect.grab but when I use the same method with jit.openni it gives out completely different results. Any ideas why this might be happening ?

jit.openni and jit.freenect.grab put out a different range of vales. You would have to scale the output of jit.openni to match the values coming from jit.freenect.grab if you wanted to drop jit.openni into code designed for freenect.grab. I don't know what that scale factor is, as I have mostly always worked with jit.openni, but I remember seeing more than one forum post about it.

If you had the sensor mounted overhead and had people below it, you would probably want to mask based on distance instead of a usermap. You could do this with a jit.expr object with an expression something like: @expr (in[0]

From above, you will likely not get any skeleton information, because the sensor has recognize a human shape in a more or less frontal view to be able to assign a user number and fit a skeleton to them.

jit.openni and jit.freenect.grab put out a different range of vales.  You would have to scale the output of jit.openni to match the values coming from jit.freenect.grab if you wanted to drop jit.openni into code designed for freenect.grab. I don't know what that scale factor is, as I have mostly always worked with jit.openni, but I remember seeing more than one forum post about it. 

If you had the sensor mounted overhead and had people below it, you would probably want to mask based on distance instead of a usermap.  You could do this with a jit.expr object with an expression something like:  @expr (in[0]

Background subtraction with jit.openni / Kinect